Enabling the Flexible Future - AMT-SYBEX...• WPD SCADA, used in Control Room , then data historian...
Transcript of Enabling the Flexible Future - AMT-SYBEX...• WPD SCADA, used in Control Room , then data historian...
Enabling the Flexible Future
15 May 2019
Agenda
• DSO transition and T.E.F. Projects
• EFFS Overview
• Forecasting
• Q&A
The Electricity System is changing
More renewables on the distribution network will displace the larger transmission connected generation.
New low carbon technologies are changing the way our customers use energy, making the system more complex and variable.
Within five years, embedded generation has come to dominate the peak power flows on distribution networks.
Embedded Generation
A traditional distribution network is designed and built to accommodate a single winter peak demand.
Our Shaping Subtransmissionreports have highlighted significant potential growth in both demand and generation, under the right political and economic environments.
5.5GW2.8GW
5.4GW3.2GW
4.0GW6.4GW
6.6GW7.8GW
WPD Shaping SubtransmissionGone Green 2030
Future Growth Scenarios
LOADGENERATION
“If we take advantage of the opportunities, we can create
new businesses and jobs, empower consumers and
help people save up to £40bn off their energy bills
in the coming decades”
Smart Systems & Flexibility Plan
Traditional DNO operations would require very substantial investments in passive grid infrastructure, which would be underutilised much of the time.
• There is an increasing risk of stranded assets or reinforcement lagging development as the growth rate of DER and LCT demand increases
• Traditional investment planning may not be able to deal with new scenarios –i.e. rapid clustering, temporary constraints, changes in diversity
• Asset replacement and traditional reinforcement will be supplemented by increasing the agility of networks and enabling customers to deliver additional flexibility when required
• Network and System operators have strong incentives to pursue smart alternatives to network upgrades where this can create value for consumers
The Need for Flexibility During Uncertainty
What is a DSO?• A Distribution Network Operator (DNO) provides a network sized to support times
of maximum demand and/or generation output. It is sufficiently large to enable the GB Market to consider it having infinite capacity.
• A Distribution System Operator (DSO) exploits ICT to deliver a network that makes optimal use of capacity
• smarter network solutions (eg. DAR; ALT, Meshing, ANM, Intertrips),
• Non-network solutions (eg. DSR, DG, Storage, Reactive Power Services).
DistributionNetworkOperator
DistributionSystemOperator
Passive networks managing maximum
power flows
Active networks managing real-time energy flows
T.E.F. Overview
T.E.F.
Network Innovation Competition (NIC) UK Context
TRANSITION, EFFS and FUSION are all electricity NIC projects. The NIC was introduced by Ofgem as part of the RIIO: ED1, T1 and GD1 price controls.
The NICs are annual competitions for electricity and gas, where network companies compete for funding for research, development and trialling for new technologies, operating and commercial arrangements.
Funding will be provided for the best innovation projects, providing a financial catalyst, which help all network operators understand what they need to do to provide environmental benefits and security of supply at value for money as Great Britain moves to a low carbon economy.
T.E.F. Project Summary
• 5 year project• Physical trials in East Fife• Based on USEF market models,
drawing on Open Network learning
• Trialling commoditised local demand-side flexibility
• 3 year project• Learning will feed into the
Cornwall Local Energy Market• Based on Open Network market
models, drawing on USEF learning
• Forecasting and data communication focused
• 5 year project• Physical trials in Oxfordshire
Based on Open Network market models, drawing on USEF learning
• Trialling of local energy flexibility and the facilitation of peer to peer trading
Project TRANSITION – Summary
Overview
TRANSITION will design, develop, demonstrate and assess the common tools, data and system architecture required to implement the proposed models produced by the Open Networks Workstream 3 project.
Benefits Work Packages
TRANSITION is focussed on implementing the outputs from Open Networks. The Neutral Market Facilitator (NMF) has the potential to deliver benefits of up to £292m for customers by 2050.
WP 1: Stakeholder Forum WP 2: Requirement design and developmentWP 3: Forecasting and DSOWP 4: Marketing Model sWP 5: IT FrameworksWP 6: Trial SpecificationWP7: Stage Gate and Trial deployment
Project
partners:
EFFS – Summary
Overview
The Electricity Flexibility and Forecasting System (EFFS) Project will specify, implement and trial a system that will to support DNO-DSO industry transition. The software will forecast network demand, generation and storage using project developed algorithms.
Benefits Workstreams
A system enabling networks to be actively managed through the use of flexibility services and delivering DSO capability that could provide DNO savings of £242.6m by 2050.
WS1: Forecasting, co-ordination & requirementsWS2: System design, development and buildWS3: Testing & trialsWS4: Collaboration & learning dissemination
Project
partners:
Project FUSION – Summary
Overview
FUSION is a 5-year (£5.67M) DSO-transition project focused on the development and implementation of a trial structured competitive market at local scale for the trading of commoditised demand side flexibility using the UniversalSmart Energy Framework (USEF). This flexibility is designed to address network constraint issues found in the distribution network, and to defer network reinforcement.
Benefits Work Packages
The FUSION Cost Benefit Analysis (CBA) has estimated the following benefits to GB electricity consumers through avoided/deferred network reinforcement:Financial Savings (NPV): £236M by 2050*(*unlocking further benefits of £3.5 Billion by 2050)Carbon Savings: 3,600 kt.CO2 by 2050
WP 1: Stakeholder Forum WP 2: Local Flexibility Quantification WP 3: USEF Fit to UK WP 4: Enabling Technologies WP 5: Live TrialWP 6: Knowledge Dissemination
Project
partners:
EFFS: Timeline
EFFS Functional Overview
Service Management
Manage & Reconcile
Scheduling
Forecasting
Defined Output
Defined Input e.g.
Historic LoadWeather Data
Capacity Engine Optimise
Reconciliation Reporting
Operations Interface
2. Arming
Dispatch Schedule
Demand Forecasting
Service ConfigurationManual Service Creation
Automatic Service Creation
Dispatch Request
Procure Request
+ 1. Procure
Procure Response
Constraints Available Flexibility
Confirmed / Purchased Flexibility
Arming Request
Arming Response
3. Dispatch
Conflict avoidance
Service data Market Interface
Forecasting
7
Requirements
• Production of load and generation forecasts for operational timescales i.e. within day, day ahead week ahead, month ahead, 3 months ahead and 6 months ahead.
• Half hourly average values - Real and Reactive power
• EHV and 132kV Transformers, feeders, directly connected load customers and generation.
• Explore weather data and impact on forecasting
• Explore machine learning vs traditional methods
• Inform conversion from Net Load to Total Load i.e. unmasking the impact of distributed generation.
Forecasting evaluation methodology and metrics
• The accuracy of the forecast values is assessed using:
• For each half hour:
The forecast tool is considered successful if:
1. Forecasts are produced for all the half-hourly periods within the specified 4 weeks.
2. The accuracy of the forecasts is found to be higher than 80% for all forecast sets for more than 80% of the time.
3. The accuracy of the forecasts does not fall below 50% for more than 50% of the time.
Accuracy Metrics (continued)
• Mean Squared Error (MSE): this is a measure of the quality of an estimator, and a value closer to zero is better.
• Mean Absolute Error (MAE): this is mean average of the absolute error between predicted and actual values. This indicated the overall error in the forecast, on average. This is used alongside the RMSE.
• Root Mean Square Error (RMSE): this is the standard deviation of the prediction errors. This is a measure of the size of the error that gives more eight to larger, or infrequent errors. The larger the difference between the MAE and RMSE, the more inconsistent the error size is.
• Mean Absolute Percentage Error (MAPE): this expresses accuracy as a percentage.
Forecasting – Data
• WPD SCADA, used in Control Room , then data historian – Data Logger. (MW and MVAr, Amps, Volts)
• Half Hourly metering data also provided (MW and MVAr)
• SCADA/metering data for significant components e.g. large customers contributing to the load at a Primary.
• Site locations – to enable matching with weather data
• Weather data
• ANM and Scottish Power load customer data.
• By Primary - Generation summary by type and capacity
• By Primary - Customer composition by profile class and combined consumption
Building and Assessing Forecasting Models
Problem Definition
Explore the Data
Model SelectionEvaluate Model
Performance
Statistical Machine Learning
ARIMA XGBoostLSTM
Few use cases in depth
More examples to achieve breadth
Forecasting – Results – GSP 6 months ahead
As expected, forecasting six months ahead has relatively large errors, likely that using weather forecasts, rather than historical weather data, would increase error significantly.
Forecasting – Results – BSP 1 month ahead
MSE MAE RMSE MAPE
0.418 0.416 0.647 4.085
Results looking far more usable.(note - we are not seeing the impact of forecast error on weather variables)
Forecasting – Results – Primary 1 day ahead
MSE MAE RMSE MAPE
0.104 0.257 0.322 4.092
Prediction looks usable and error reduced compared to month ahead.
Impact of scale as well as time horizon. BSP has more customers than Primary so more averaging of behaviour, but this may be offset by greater capacity of DG connected.
Forecasting – Findings to date
• Machine learning methods results are better than ARIMA overallXGBoost provided the best balance between accuracy and complexity of operation.
• Need to develop and tune models for each location. Ongoing requirement to keep monitoring forecast performance and retrain.
• May get some indication of the best model types for site types, but this may only emerge after extending the work for EFFS trial.
• GSP forecast improved by forecasting each transformer separately.
Forecasting – Findings to date (continued)
• BSP and Primary forecasts can be acceptable without weather data, though including these features did improve the models.
• Renewable generation - better to use engineering model than time series forecasting method due to the non-linear relationship (Renewables Ninja website very helpful)
• 6 months of historic data is adequate to train forecasts for a month ahead or shorter time horizons – better than when 12 months of data was used.
• SGS report to be published in June
Solution
Postgres database to host time-series and spatial datasets for training, testing andevaluation.
Anaconda data science platform to access development environments, forecastingmethods and data visualisation tools.
Python workbooks to manage data import and export from Anaconda.
Open source toolchain toaccess the broadest range offorecasting methods andensure repeatability forother DNOs.
EFFS trials
• Many more sites – at least 200 locations Should be able to extend SGS analysis to help select best model parameters according to forecast site type
• Extend to 11kV Primary Interconnectors
• Weather forecast rather than historic data
• Share output with National Grid ESO
• Investigate options for automation of
– Data provision
– Model training
– Model assessment
– Model retraining
Q&A
Any questions?
Contact
Jenny Woodruff (Project Manager)
Elliot Warburton (Delivery Manager)